How to assign numerical values to individual welfare?post by Frank_R · 2021-07-29T13:52:19.322Z · EA · GW · 2 comments
This is a question post.
I have the following question about utilitarianism. If you define
a utility function u, you have to sum over the welfare of all
sentinent beings. It is more or less possible to decide if
the welfare of one being is greater than the welfare of another being.
In order to determine the total utility an order relation is not
enough. In addition you have to assign a numerical value
to the individual welfare. Depending on how you choose
this value the best decisions can be vastly different.
To illustrate my point I introduce the following toy model: Let
there be n classes of sentinent beings whose mental
capacibilities are comparable. The first class could be insects,
the second rodents, the third apes, the fourth humans
and the fifth AIs with superhuman mental capacibilities.
I denote the set of all happy beings in the class i by
and the set of all suffering beings by . The
number of all beings in should be .
By observing the behaviour or the brain architecture
of the beings, we can get a rough idea if individuals in
have higher cognitive functions / a higher
welfare than members of (although even this may be
debatable). Mathematically speaking we have an order relation
on . In order to determine the total
utility u we need a monotonous function
If f has large negative values for negative indices, we
obtain suffering focused ethics. If f is nearly constant
for positive indices, the ideal world would be populated only
by happy insects, since a fixed area can support more of
them. If f is nearly zero for non-human animals and 1
for humans and AIs, the ideal world would be populated by
as many happy humans (or emulated minds) as possible
and AI development would not be a high priority. If f
is increasing very fast, it would be the best to develop
superhuman AIs as fast as possible even at the expense of humans.
My question is if there is a non-arbitrary way to determine
f. It may be possible to define f in such a way
that it captures most of ethical intuitions. I am not sure
if this approach is to subjective, but I do not see how
to define f in terms that can be measured.
Comments sorted by top scores.